System and method for identifying objects of interest in images based on likelihood map decluttering

An automatic threat recognition system for scanning an article to identify different objects of interest (OOI) contained within the article, wherein the system makes use of a computed tomography (CT) scanner for scanning the article to obtain a CT image scan of the objects within the article, the reconstructed CT image scan represented by a plurality of two dimensional (2D) image slices of the article and its contents, the system comprising:

• a computer which receives each said (2D) image slice and obtains therefrom information forming a plurality of voxels representing each said image slice;
• the computer configured to use the voxels associated with each said CT image slice, in connection with a machine learning operation, to:
• analyze the voxels to create likelihood maps, one for each MOI, representing likelihoods that voxels making up the CT image are associated with each one of said materials of interest (MOI);
• for each likelihood map, the computer is configured to:
• analyze the voxels in the likelihood map by using moving average algorithms, in three dimensions, and computing engineered voxel neighborhood features, independent of cost and neighborhood size;
• train a binary classification tree (BCT) to learn the neighborhood size and feature thresholds that jointly minimize discrepancy between a decluttered likelihood map and a ground truth map for the MOI;
• declutter the likelihood maps by evaluating trained BCTs on vectors of likelihood map voxel neighborhood features;
• and
• the computer further being configured to merge the decluttered likelihood maps for the MOIs into a single disambiguated material map, and to connect voxels of a like MOI together to form segments, and to connect segments together to construct candidate OOIs to assist with rapid identification of the OOIs.

A method for identifying an object of interest (OOI) within an article, in an automated fashion, the method comprising:

• using a computed tomography (CT) scanner to scan the article to obtain a CT image scan of the objects within the article, the reconstructed CT image scan represented by a plurality of two dimensional (2D) image slices of the article and its contents, wherein each said 2D image slice includes information forming a plurality of voxels;
• using a computer configured to receive the information and, for the voxels associated with the CT image, to use machine learning to:
• analyze the voxels to create a likelihood map representing likelihoods that voxels making up the CT image are associated with at least one material of interest (MOI);
• and
• analyzing the likelihood map to construct neighborhoods of voxels within the likelihood map, and using evaluations of the neighborhoods of voxels to declutter the likelihood map, to connect voxels of a like MOI together into segments, and to connect the segment of a like MOI together to help construct candidate OOIs to help enable rapid identification of OOIs.

An automatic inspection system for scanning a workpiece article to identify at least one of a void or crack in the workpiece, and where the system makes use of a computed tomography (CT) scanner for scanning the workpiece to produce a reconstructed CT image scan made up of a plurality of two dimensional (2D) image slices of the workpiece, the system comprising:

• a computer which operates to obtain, from each said (2D) image slice, information forming a plurality of voxels making up the image slice;
• the computer configured to receive the information and, for the voxels associated with the CT image, to use machine learning to:
• analyze the voxels to create a likelihood map representing likelihoods that voxels making up the CT image are associated with a material of interest (MOI);
• and
• further analyzing the likelihood map to construct neighborhoods of voxels within the likelihood map, and using evaluations of the neighborhoods of voxels to declutter the likelihood map, to connect voxels associated with the defect together to form segments, wherein the segments indicate at least one of a void or a crack in the workpiece.